Refine
Document Type
- Article (3)
Language
- English (3)
Has Fulltext
- yes (3)
Is part of the Bibliography
- no (3)
Keywords
- Machine learning (1)
- Medical imaging (1)
- Pathology (1)
- Prostate cancer (1)
Institute
- Medizin (3)
The Gleason grading system remains the most powerful prognostic predictor for patients with prostate cancer since the 1960s. Its application requires highly-trained pathologists, is tedious and yet suffers from limited inter-pathologist reproducibility, especially for the intermediate Gleason score 7. Automated annotation procedures constitute a viable solution to remedy these limitations. In this study, we present a deep learning approach for automated Gleason grading of prostate cancer tissue microarrays with Hematoxylin and Eosin (H&E) staining. Our system was trained using detailed Gleason annotations on a discovery cohort of 641 patients and was then evaluated on an independent test cohort of 245 patients annotated by two pathologists. On the test cohort, the inter-annotator agreements between the model and each pathologist, quantified via Cohen’s quadratic kappa statistic, were 0.75 and 0.71 respectively, comparable with the inter-pathologist agreement (kappa = 0.71). Furthermore, the model’s Gleason score assignments achieved pathology expert-level stratification of patients into prognostically distinct groups, on the basis of disease-specific survival data available for the test cohort. Overall, our study shows promising results regarding the applicability of deep learning-based solutions towards more objective and reproducible prostate cancer grading, especially for cases with heterogeneous Gleason patterns.
Background: Intracerebral haemorrhage growth is associated with poor clinical outcome and is a therapeutic target for improving outcome. We aimed to determine the absolute risk and predictors of intracerebral haemorrhage growth, develop and validate prediction models, and evaluate the added value of CT angiography.
Methods: In a systematic review of OVID MEDLINE—with additional hand-searching of relevant studies' bibliographies— from Jan 1, 1970, to Dec 31, 2015, we identified observational cohorts and randomised trials with repeat scanning protocols that included at least ten patients with acute intracerebral haemorrhage. We sought individual patient-level data from corresponding authors for patients aged 18 years or older with data available from brain imaging initially done 0·5–24 h and repeated fewer than 6 days after symptom onset, who had baseline intracerebral haemorrhage volume of less than 150 mL, and did not undergo acute treatment that might reduce intracerebral haemorrhage volume. We estimated the absolute risk and predictors of the primary outcome of intracerebral haemorrhage growth (defined as >6 mL increase in intracerebral haemorrhage volume on repeat imaging) using multivariable logistic regression models in development and validation cohorts in four subgroups of patients, using a hierarchical approach: patients not taking anticoagulant therapy at intracerebral haemorrhage onset (who constituted the largest subgroup), patients taking anticoagulant therapy at intracerebral haemorrhage onset, patients from cohorts that included at least some patients taking anticoagulant therapy at intracerebral haemorrhage onset, and patients for whom both information about anticoagulant therapy at intracerebral haemorrhage onset and spot sign on acute CT angiography were known.
Findings: Of 4191 studies identified, 77 were eligible for inclusion. Overall, 36 (47%) cohorts provided data on 5435 eligible patients. 5076 of these patients were not taking anticoagulant therapy at symptom onset (median age 67 years, IQR 56–76), of whom 1009 (20%) had intracerebral haemorrhage growth. Multivariable models of patients with data on antiplatelet therapy use, data on anticoagulant therapy use, and assessment of CT angiography spot sign at symptom onset showed that time from symptom onset to baseline imaging (odds ratio 0·50, 95% CI 0·36–0·70; p<0·0001), intracerebral haemorrhage volume on baseline imaging (7·18, 4·46–11·60; p<0·0001), antiplatelet use (1·68, 1·06–2·66; p=0·026), and anticoagulant use (3·48, 1·96–6·16; p<0·0001) were independent predictors of intracerebral haemorrhage growth (C-index 0·78, 95% CI 0·75–0·82). Addition of CT angiography spot sign (odds ratio 4·46, 95% CI 2·95–6·75; p<0·0001) to the model increased the C-index by 0·05 (95% CI 0·03–0·07).
Interpretation: In this large patient-level meta-analysis, models using four or five predictors had acceptable to good discrimination. These models could inform the location and frequency of observations on patients in clinical practice, explain treatment effects in prior randomised trials, and guide the design of future trials.
Funding: UK Medical Research Council and British Heart Foundation.